Chapter 4: Tool Use - Read Pre-Recon Data

Welcome back! In the previous chapter, Tool Use - Read Exploitation Queue, our agent learned how to pick up its "To-Do List" of specific attacks.

However, a good detective doesn't just blindly follow a list. They need to understand the context. Before we start breaking into things, we need to read the initial background reports.

Why do we need to Read Pre-Recon Data?

Imagine you are that detective again. You have a list of suspects to interview (The Exploitation Queue), but before you talk to them, you should probably read the initial police report filed by the first officer on the scene.

Pre-Recon Data is that initial police report. It stands for "Preliminary Reconnaissance." It contains high-level information gathered before the deep investigation started.

The Use Case

Our agent is targeting http://localhost:33081. While the "Queue" tells the agent where to attack, the Pre-Recon Data might contain notes about what technologies the website uses (e.g., "This site uses PHP" or "The server is Apache").

The agent reads the file deliverables/pre_recon_deliverable.md to load this background context into its brain.

Key Concepts

1. Context: General information that helps the agent make better decisions. Knowing the website uses "PHP" might make the agent try PHP-specific attacks later.
2. Markdown (.md): Unlike the previous chapter where we read a JSON list, this data is stored in a Markdown file. This is a text format used for writing human-readable reports with headers and bullet points.
3. Unstructured Data: The "Queue" was structured (strict rows and columns). "Pre-Recon" is unstructured (paragraphs and notes). The agent reads this differently.

How to Use the Tool

Reading a text file is one of the most fundamental skills for an agent.

Step 1: Define the File Path

We tell the agent where the initial report is located.

# The path to the preliminary report
report_path = "deliverables/pre_recon_deliverable.md"

print(f"Reading background info from: {report_path}")

Output: Reading background info from: deliverables/pre_recon_deliverable.md

Step 2: Read the File

We use the agent's file-reading tool. Notice that we are just reading raw text this time, not parsing a list.

# The agent uses 'read_file' to get the text content
background_info = agent.tools.read_file(report_path)

# Let's peek at the first 50 characters
print(f"Report Start: {background_info[:50]}...")

Output:

Report Start: # Initial Scan Results
Target: http://localhost:330...

The agent has now absorbed the report. It holds the text in its memory, ready to be analyzed or sent to a Large Language Model (LLM) to extract meaning.

Under the Hood: What happens?

How does the agent physically open and read this report?

The Workflow

This process is very similar to opening a document in Notepad, but it happens instantaneously via code.

Internal Implementation

Let's look at the code inside shannon/tools/file_reader.py. This is even simpler than the JSON reader we saw in the last chapter.

class FileTool:
    def read_file(self, file_path):
        # 1. Open the file in 'read' mode
        with open(file_path, 'r', encoding='utf-8') as f:
            
            # 2. Read the entire file into a single string
            content = f.read()
            
        # 3. Return the text
        return content

Explanation:

1. encoding='utf-8': This ensures that if the report contains special characters (like emojis or foreign languages), the agent reads them correctly without crashing.
2. f.read(): This command grabs everything from the first letter to the very last period and stores it as one giant piece of text.

What's Next?

Our agent is getting smarter. It has a list of tasks (from the Queue) and it has read the initial police report (Pre-Recon Data).

But there is one massive piece of evidence missing. We need the Detailed Map of the target—the deep reconnaissance data that lists every single page, image, and link on the website.

In the next chapter, we will learn how the agent ingests this massive dataset.

Next Chapter: Tool Use - Read Recon Data

Generated by Code IQ